As more targeted therapies are approved for different types of cancer, there is a growing need for predictive biomarkers so that these therapies can be directed to patients who will most benefit from them; unfortunately, the biomarkers available for cancer therapy are not sufficient. Recently, the development of tyrosine kinase inhibitors (TKI) improved treatment in patients with advanced disease. For example, detection of mutations in EGFR has been successfully used as a biomarker for initial therapy with EGFR inhibitors. Many targeted agents lack genetic predictive biomarkers. Furthermore, resistance to these drugs frequently emerges, and it is often not clear what treatment is best given following this emergence of resistance, given the variety of mechanisms for resistance. The present invention provides a method of predicting the response to therapy so that drugs can better be assigned to patients.